EP0355506B1 - Arrangement for measuring local bioelectric currents in biological tissue - Google Patents

Description

The invention relates to an arrangement for measuring local bioelectric currents in biological tissue complexes according to the preamble of claim 1. Such an arrangement is known from US Pat. No. 4,736,751. It is used to analyze brain activity on a statistical basis with the help of a digital computer. For this purpose, sensors are distributed in the usual way over the skull, the measurement signals of which are digitized and stored. The patient to be examined is exposed to a set of stimulating sensory stimuli that trigger electrical or magnetic fields in localized areas of the brain. These fields are converted into electrical signals by the sensors, which are statistically evaluated and analyzed in order to verify the respective location of the activity. Due to the spatial and temporal relationships of these spontaneous events, the location of their origin and their spread can be shown on a three-dimensional skull model, produced for example by a computer tomogram using the magnetic resonance method (MR).

The disadvantage of this method is that it only leads to useful results if the evoked potentials are large enough to stand out clearly from the noise level. This generally applies to evoked potentials after a larger number of stimuli and their averaging. However, the method is not suitable for the determination of so-called spontaneous events, such as those that occur during an epileptic attack. According to the current state of knowledge in the electroencephalogram (EEG), these are characterized by characteristic signal patterns, e.g. so-called "Spike and wave complexes" of about 200 to 500 ms duration noticeable. These signal patterns can also be determined between acute attacks (interictal), but with a very different frequency from patient to patient. In extreme cases, such signal patterns can occur every second or only once a week. Due to the low signal-to-noise ratio, such interictal signal patterns are usually difficult to recognize in the EEG and only practically impossible to see with the naked eye in the magnetoencephalogram (MEG) by experienced neurologists. The origin of such spontaneously occurring signal patterns is interpreted as an epileptogenic focus. The aim of evaluating an EEG or MEG in epilepsy diagnostics is to localize the location of this focus as precisely as possible. It is also important for the neurologist to make statements about the spatial spread of signal-forming electrical excitations both within a signal pattern and in the case of successive different signal patterns. Such statements have so far only been possible invasively using EEG depth electrodes and are therefore also only possible to a very limited extent. A time-resolved localization has so far been practically impossible to achieve, since due to the low signal-to-noise ratio from a single signal event, localization with the required accuracy is not possible and, as a rule, there were not enough such events available for averaging signal events.

The invention is based on the object of processing the electrical signals emanating from sensors arranged spatially distributed over the biological tissue complex to be examined for the measurement of electrical or magnetic field strength values in such a way that certain signal patterns occurring within a time interval form the basis for the search for similar patterns can serve the current signal and determine their temporal occurrence and their spatial allocation.

This object is achieved by the invention specified in patent claims 1 and 2. The spatial and temporal correlation of the averaged signal pattern proposed therein shows usable signal values that stand out sufficiently from the noise and can be evaluated accordingly.

In a development of the invention according to claim 3 it is achieved that evaluable interference frequencies, such as the network frequency or certain periodically occurring spontaneous biosignals can be filtered out of the signal channels of the sensors.

According to a further embodiment according to claim 4, the spatial correlation is formed by averaging the correlation function over the assigned time interval of the signal pattern and the temporal correlation function with subsequent averaging over the assigned space according to certain mathematical relationships. Through their subsequent multiplication, peak signals can be obtained that stand out particularly clearly from the noise.

The arrangement described in claim 2 enables the determination of the similarity threshold characteristic of a measurement data set.

The required arithmetic operations, when carried out by the usual digital computers, take up a considerable amount of computing time. In a further development of the invention according to claim 7, the use of an array processor computer in conjunction with an algorithm for so-called fast folding enables a significant reduction in the computing time to be achieved, which allows results to be obtained shortly after the examination, so that the patient can be used for a possible repetition can remain connected to the measuring arrangement.

It is known from "Biogagnetism" 1987, "Proceedings of the 6th International Conference on Biomagnetism" under the title "New Method for the Study of Spontaneous Brain Activity" for the detection of epileptic and alpha activities, a correlation of the spatial patterns for recognition of "spike wave complexes" in only one EEG channel. However, this method is only sufficient if a significant event can be clearly recognized in a channel. However, it has been shown that in the context of the signal-to-noise ratios that occur in practice, significant correlations of "spike-wave complexes" cannot be observed at all in the simple correlation function and only very poorly in the spatially and temporally averaged correlation functions of an MEG are. Only by treating the signal complexes according to the present invention do the characteristic signal patterns stand out so clearly from the noise that they can be used for further signal treatment and signal evaluation.

FIG 1

eine Meßanordnung für die Behandlung der von den am Patienten angeordneten Sensoren gelieferten Meßwerte;

FIG 2

ein typisches Signalmuster eines spontanen Ereignisses im Hirnbereich eines Patienten;

FIG 3

das nach einer Raum-zeitlichen Korrelation erhaltene gemittelte Signal gemäß FIG 2;

FIG 4

eine typische Verteilungskurve, die die Häufigkeit des Auftretens eines bestimmten Ähnlichkeitsmaßes (Korrelationskoeffizienten) zwischen der Signalaufzeichnung und dem Vergleichssignal darstellt;

FIG 5

eine Schaltungsanordnung zur Bestimmung der Ähnlichkeitsschwelle; und

FIG 6

das von einem Computertomographen gewonnene Schichtbild des Gehirns mit einer eingezeichneten typischen Erregungsbahn eines spontanen Ereignisses.

Further details of the invention result from the following explanations of an exemplary embodiment shown in FIGS. 1 to 4. In it show:

FIG. 1

a measuring arrangement for the treatment of the measured values supplied by the sensors arranged on the patient;

FIG 2

a typical signal pattern of a spontaneous event in the brain area of a patient;

FIG 3

the averaged signal obtained after a space-time correlation according to FIG. 2;

FIG 4

a typical distribution curve representing the frequency of the occurrence of a certain measure of similarity (correlation coefficient) between the signal recording and the comparison signal;

FIG 5

a circuit arrangement for determining the similarity threshold; and

FIG 6

the slice image of the brain obtained by a computer tomograph with a typical path of excitation of a spontaneous event.

In FIG. 1, the sensor electrodes 1 of an electroencephalograph (EEG) and the squid sensor of a multi-channel magnetoencephalograph (MEG) 2 are arranged spatially distributed over the skull cap 3 of a patient. The sensors generate electrical signals corresponding to the measured electrical or magnetic fields, which are fed via leads 4, 5 to an N-channel A / D converter 6. Furthermore, the A / D converter can trigger signals from an EKG device via a feed line 7 and trigger signals controlled by breathing can be fed via the feed line 8, which serve in a known manner to determine the measured value acquisition within certain respiratory rate and / or cardiac activity trigger time limits. The digital signals of the EEG and MEG channels are fed to an N-channel digital frequency filter 9, which filters out periodically occurring interference frequencies, such as the network frequency or excitation centers of the so-called alpha waves emanating from the brain. The output of the frequency filter 9 is connected to an EEG monitor 10, which displays the output signals in an evaluable form (digital or analog) and thus makes them available for analysis by the doctor.

Alternatively, it is possible to provide a freely programmable pattern recognition stage 11 instead of the EEG monitor 10, which can be programmed by means of a programming device 12 for the recognition of certain signal patterns.

The recognized signal pattern must then be defined in time. For this purpose, a start time and an end time are defined. A Such a time-defined signal pattern is referred to as a "template". The template recognized and defined in this way is stored in the template storage stage 13.

Ebenso wird der Korrelationskoeffizient der Signalkurven in dem betreffenden Fenster mit Hilfe eines zweiten Rechenschaltkreises 25 nach der folgenden Formel

an jedem Meßort berechnet und über alle Zeitpunkte des Fensters gemittelt. Die so berechneten Zeitfunktionen der zeitlichen und räumlichen Korrelation werden anschließend mit Hilfe eines dritten Rechenschaltkreises 26 nach der Formel

${\text{K}}_{\text{RT}}{\text{(t}}_{\text{i}}{\text{) = K}}_{\text{R}}{\text{(t}}_{\text{i}}{\text{) . K}}_{\text{T}}{\text{(t}}_{\text{i}}\text{)}$

multipliziert.The continuously measured signal at the output of the digital frequency filter 9 is fed to a correlation stage 14 which can also call up the template stored in the template storage stage 13 in order to compare it with the continuously incoming signal. To do this, the time interval defined as a template is pushed over the incoming data as a "time window". In each time window, the correlation coefficient of each signal pattern corresponding in time is calculated with the aid of its first arithmetic circuit 24 according to the following mathematical relationship and averaged over all measurement locations:

Likewise, the correlation coefficient of the signal curves in the window in question is determined using a second arithmetic circuit 25 according to the following formula

calculated at each measuring location and averaged over all times of the window. The time functions of the temporal and spatial correlation calculated in this way are then calculated using a third computing circuit 26 according to the formula

${\text{K}}_{\text{RT}}{\text{(t}}_{\text{i}}{\text{) = K}}_{\text{R}}{\text{(t}}_{\text{i}}{\text{). K}}_{\text{T}}{\text{(t}}_{\text{i}}\text{)}$

multiplied.

magnetisches Signal im Kanal C_i zum Zeitpunkt t_i
τ₀ ... τ_i ... τ_M Zeitintervall des Templates, Beginn τ_o, Ende τ_M, mit M Abtastwerten im Zeitintervall;
τ_i Zeitpunkt im Zeitintervall τ_o ≦ τ_i ≦ τ_M
K_T (t_i) zeitlicher Korrelationsfaktor, (Korrelationskoeffizient zum Zeitpunkt t_i von Template und Meßsignal)
K_R (t_i) räumlicher Korrelationsfaktor (Korrelationskoeffizient zum Zeitpunkt t_i von Template und Meßsignal)
K_RT (t_i) Raum-zeitlicher Korrelationsfaktor
Das so erhaltene, nach der korrelierten Raum-Zeit-Funktion gebildete Korrelationssignal wird einer Vergleichsstufe 16 zugeführt, welche dieses Korrelationssignal mit einem Schwellwertsignal vergleicht, das in einer Schwellwertdefinitionsstufe 15 aus dem Ausgangssignal des Frequenzfilters 9 gewonnen wird. Bei Überschreiten der Schwelle wird das Schwellwertsignal einer Mittelungsstufe 17 zugeführt. Diese bildet ein Mittelwertsignal aller aus den vorgenannten Kriterien erkannter Signalmuster über der Zeit an allen Meßorten, welches einerseits der Template-Speicherstufe 13 und andererseits der Schwellendefinitionsstufe 15 zur laufenden Korrektur der Templates zugeführt ist. Darüber hinaus werden die von der Mittelungsstufe 17 gemittelten Signalmuster einer Lokalisierungsstufe 18 zugeführt, welche den geometrischen Ort der aufgetretenen pathologischen elektrisch aktiven Quelle berechnet und diese Daten einer Koordinatentransformationsstufe 19 zuführt, die das Koordinatensystem der EEG-bzw. MEG-Messung mit demjenigen, beispielsweise eines in der Bildspeicherstufe 20 gespeicherten Computertomogramms, zur Deckung bringt, so daß beide Darstellungen auf einem Monitor 21 als Schnittbild oder als räumliches Bild zur Darstellung gebracht werden können.Where:
C₁ ... C _N N magnetic measuring channels in any local distribution

magnetic signal in channel C _i at time t _i
τ₀ ... τ _i ... τ _M time interval of the template, start τ _o , end τ _M , with M samples in the time interval;
τ _i time in the time interval τ _o ≦ τ _i ≦ τ _M
K _T (t _i ) temporal correlation factor, (correlation coefficient at time t _i of template and measurement signal)
K _R (t _i ) spatial correlation factor (correlation coefficient at time t _i of template and measurement signal)
K _RT (t _i ) spatio-temporal correlation factor
The correlation signal obtained in this way, formed according to the correlated space-time function, is fed to a comparison stage 16, which compares this correlation signal with a threshold value signal, which is generated in a threshold value definition stage 15 from the output signal of the frequency filter 9 is obtained. If the threshold is exceeded, the threshold value signal is fed to an averaging stage 17. This forms an average value signal of all signal patterns recognized from the aforementioned criteria over time at all measuring locations, which is supplied on the one hand to the template storage stage 13 and on the other hand to the threshold definition stage 15 for the ongoing correction of the templates. In addition, the signal pattern averaged by the averaging stage 17 is fed to a localization stage 18, which calculates the geometric location of the pathological, electrically active source that has occurred and feeds this data to a coordinate transformation stage 19, which the coordinate system of the EEG or MEG measurement coincides with that, for example a computer tomogram stored in the image storage stage 20, so that both representations can be displayed on a monitor 21 as a sectional image or as a spatial image.

An alternative method of finding very weak signal patterns, which leads to better results from case to case, consists in that the sum amounts of the signals of all channels are formed with the aid of an amount sum stage 22 and are fed to a pattern recognition stage 23 which corresponds to the pattern recognition stage 11, 12. Instead of the signal coming from the comparison stage 16, the recognized pattern is fed to the averaging stage 17 and processed in the manner described. The sum signal can alternatively also be supplied to the pattern recognition stage 11, 12 and used for the definition of the template. This type of signal processing is particularly suitable for signal generation only via MEG sampling.

FIG. 2 shows an EEG curve of the signal curve generated by an EEG electrode, in which triangular signals S1 to S9, which are referred to by the neurologist as "Sharp-Wave", are noticeable and whose pathological significance is apparent however, is not clear. The complex S₂, S₃ stands out from the rest. This was therefore used as a template and is shown hatched.

In FIG. 3, the averaged signal obtained after the spatio-temporal correlation is shown hatched in the same EEG channel. The signal pattern averaged in this way fulfills the criteria of a pathological "spike-wave complex" considerably more clearly, but shows a more complicated structure than was represented in the previously known EEG.

The arrangement described so far allows the detection of certain events from a continuous recording of bioelectric or biomagnetic signals by means of a digital spatio-temporal correlation analysis by comparing the continuously arriving signal recording with a stored defined signal pattern (template).

As a result of this comparison, a value between -1 and +1 is obtained in the data set for the correlation coefficient at each comparison time, which is a measure of the similarity of the signal recording within the time window determined by the signal pattern at each comparison time. If the correlation coefficient is +1, the agreement under the same sign is maximal. The correlation coefficient reaches the worst match at zero and a maximum match at -1, with the opposite sign of the signals. The aim of the further development of the invention is not only to find out from the signal sequence those signal areas which are identical to the template, but also to have a characteristic degree of similarity. Those signal areas are to be registered which exceed a similarity threshold which is characteristic of the relevant data record.

The frequency of the occurrence of all possible similarity measures between the signal recording and the comparison signal is shown in FIG. 4 using a typical distribution curve. If the examined signal area consists only of white noise, the frequency distribution of all correlation coefficients represents a Gaussian normal distribution, as shown as a dashed curve N in FIG. 4. Any deviation of the frequency distribution, the so-called histogram, from the normal distribution, as shown for example in the solid curve H in Fig. 4, is a clear sign that there are signal complexes which, depending on the size of their respective correlation coefficient, one are more or less similar to the given curve shape of the template. Such deviations are expressed in peaks P₁ ... P₈, which overlay the normal distribution curve N. The closer such a peak is to +1, the greater the degree of similarity. As a result, the base point on the left (i.e. in the direction of small correlation coefficients) of the tip that is closest to the correlation coefficient +1 determines the characteristic similarity threshold sought. In the present case, the similarity threshold sought would be determined from the peak P₈ with 0.48. Each exceeding of this threshold value defines a point in time in the examined signal that is sufficiently similar to the template.

A circuit arrangement for determining the similarity threshold is shown in FIG. 5. Those stages which have the same function as those in FIG. 1 are provided with the same reference symbols.

The difference from the circuit arrangement shown in FIG. 1 is that the measurement signal in the memory stage 27 and the correlation signal formed by the correlation stage 14 is stored in a storage stage 28. This signal is fed to a computing stage 29 for computing the histogram and at the same time to the comparison stage 16. The histogram signal occurring at the output of the computing stage 29 is fed to a threshold value determination stage 30, which determines the characteristic threshold value from the distribution curve and also feeds it to the comparison stage 16. In the comparison stage 16, the stored correlation signal from the storage stage 28 is compared with the characteristic threshold value from the threshold value determination stage 30 and, if the threshold value is exceeded, the signal section belonging to this time is fed to the averaging stage 17 and from there evaluated via the localization stage 18 as already described. In this way, signals are also detected for evaluation that exceed a characteristic similarity threshold to the template signal, so that even if the noise amplitude is unknown or if the total signal from the searched signal, noise and other characteristic signal complexes is recognized, the sought-after signal is recognized Signal complexes with subsequent averaging is possible.

FIG. 6 shows the image recognizable on the monitor 20, in which the computer tomogram is made to coincide with the coordinate-transformed MEG localization image. From this, the area of the pathological electrical activity through the points marked with crosses can be clearly recognized by a line characterizing the temporal activity course.

Claims (7)

1. An arrangement for localizing bioelectrical current sources in biological tissue complexes with sensors arranged spatially distributed over said tissue complexes for the formation of electrical measured values on the basis of measured electrical or magnetic field quantities produced by the bioelectrical current sources with

   - a stage (17) for the averaging of similar signal patterns,

   - with a stage (18) for the spatial assignment of the averaged signal patterns to the signal-triggering current sources and

   - with a stage (21) for the spatial representation of the current sources within an image of the tissue complex, characterised by a signal processing arrangement consisting of a stage (11, 12, 13) for visual or automatic recognition of individual signal patterns present in the measured values, temporal limitation of the signal pattern (time window), and storage of the measured values, present in the time window, from each sensor (template), a further stage (14) for spatial and temporal correlation of the continuous measured values of each sensor with the signal pattern present within the time window and assigned to the respective sensor for the formation of a correlation signal, a threshold value definition stage (15) which produces a threshold value from the continuous measured values, a comparator stage (16) which compares the correlation signal with the threshold value and on the overshooting of the threshold value supplies continuous measured values, from which the correlation signal overshooting the threshold value has been formed, to the averaging stage (17).
2. An arrangement for the localization of bioelectrical current sources in biological tissue complexes with sensors arranged spatially distributed over said tissue complexes for the formation of electrical measured values on the basis of measured electrical or magnetic field quantities produced by the bioelectrical current sources with

   - a stage (17) for the averaging of similar signal patterns,

   - with a stage (18) for the spatial assignment of the averaged signal patterns to the signal-triggering current sources and

   - with a stage (21) for the spatial representation of the current sources within an image of the tissue complex, characterised by a signal processing arrangement consisting of a stage (11, 12, 13) for visual or automatic recognition of individual signal patterns present in the measured values, temporal limitation of the signal pattern (time window), and storage of the measured values present in the time window from each sensor (template), a further stage (14) for spatial and temporal correlation of the continuous measured values of each sensor with the signal pattern which is present within the time window and is assigned to the respective sensor for the formation of a correlation signal, a storage stage (28) for the intermediate storage of the correlation signal and with a calculating stage (19) for the calculation of a frequency distribution (histogram) of the stored correlation signal, a threshold value determining stage (30) which, on the basis of a defined deviation of the frequency distribution from a Gaussian normal distribution curve, establishes a similarity threshold characteristic of the investigated signal portion and supplies the latter to a comparator stage (16) which compares the intermediately stored correlation signal with the similarity threshold value and in the event of the overshooting of the threshold value supplies continuous measured values, from which the correlation signal overshooting the threshold value has been formed, from the storage stage (27) to the averaging stage (17).
3. An arrangement as claimed in Claim 1 or 2, characterised in that the electrical signal patterns formed by the sensors (1, 2) are digitalized by an analogue-digital converter (6) arranged at the output end of the sensors (1, 2) and are fed to a digital frequency filter (9) which filters out selectable interference frequencies from the signal channels (4, 5).
4. An arrangement as claimed in one of Claims 1 to 3, characterised in that the stage (14) for the spatial and temporal correlation is formed from a first calculating circuit (24) for the formation of the temporal correlation function and its averaging over the assigned space in accordance with the formula

   

   a second circuit (25) for the formation of the spatial correlation function with subsequent averaging over the assigned time window in accordance with the formula

   

   and a third calculating circuit (26) for the multiplication of the two functions, where the formula symbols have the following significance:
   C₁ ...C_N N magnetic measuring channels in arbitrary local distribution

   

   magnetic signal in the channel C_i at the time t_i
   τ₀...τ_i...τ_M time interval of the template, start τ_o, end τ_M, with M sample values in the time interval; τ_i time in the time interval τ_o ≦ τ_i ≦ τ_M
   K_T (t_i) temporal correlation factor, (correlation coefficient at the time t_i of template and measured signal)
   K_R (t_i) spatial correlation factor (correlation coefficient at the time t_i of template and measured signal)
   K_RT (t_i) space-time correlation factor.
5. An arrangement as claimed in Claim 1 or 2, characterised in that the stage (14) for the spatial and temporal correlation comprises one single calculating circuit which undertakes the correlation in accordance with the formula

   

   where the formula symbols have the significance defined in Claim 4.
6. An arrangement as claimed in one of Claims 3 to 5 in association with Claim 1, characterised in that the average signal formed by the averaging stage (17) is fed to the template storage stage (13) and/or to the threshold value definition stage (15) for the continuous correction of the templates.
7. An arrangement as claimed in one of the preceding claims, characterised in that an array processor computer is used for the formation of the space-time correlation functions, and that the correlation function is formed using a "fast convolution" algorithm.

Images